Abstract 7431: IAMSAM: Image-based analysis of molecular signatures using the Segment-anything model - Integrative analysis tool for tumor microenvironment

Abstract

Abstract Introduction. Spatial transcriptomics (ST) is a powerful approach for investigating gene expression patterns in tissues while preserving their spatial context. However, working with ST data presents challenges due to its complexity, high dimensionality, and the absence of user-friendly tools. IAMSAM (Image-based Analysis of Molecular signatures using the Segment-Anything Model) leverages the Segment-anything model (SAM), a cutting-edge deep learning model developed by Meta, which is specifically designed for image segmentation. We demonstrated IAMSAM's capabilities with publicly available ST datasets, showcasing its user-friendly interface, featuring its user-friendly interface for exploring pathophysiology and potential biomarkers. Methods. IAMSAM is a web-based tool designed for analyzing ST data, utilizing the SAM for image segmentation. It offers two modes: everything-mode and prompt-mode. In everything-mode, the model automatically generates segmentation masks for the entire histologic image. In prompt-mode, users can draw rectangle boxes to provide input prompts for SAM. IAMSAM's downstream analysis includes the identification of DEGs, enrichment analysis, and cell type prediction of selected regions of interest. DEGs are represented in volcano plots and box plots, aiding in the identification of genes associated with image features of tumors such as cancer-enriched areas or stromal patterns. Enrichment analysis assesses gene sets for over-representation, providing insights into pathways and biological processes relevant to cancer. Cell type prediction utilizes the CellTypist API, offering valuable information about the cellular composition and heterogeneity within the tumor microenvironment. Results. IAMSAM has been applied to publicly available ST datasets to demonstrate its effectiveness in cancer research. It accurately identified cancer markers in breast cancer samples and aligned with spot clustering results in various cancer samples, all of which have significant implications for understanding cancer biology. Prompt-mode effectively identified stromal regions in tumor sections, and with magnification, it uncovered microvessels in prostate cancer samples, providing valuable insights for cancer pathophysiology research. IAMSAM was also applied to fluorescence imaging, demonstrating its versatility and potential in various cancer-related studies. Conclusions. IAMSAM is a valuable web-based tool for cancer research using spatial transcriptomics data particularly integratively analyzing tumor microenvironments with a user-friendly interface. IAMSAM opens up new avenues for spatial transcriptomics research in the context of cancer, making it a promising resource for cancer researchers. Citation Format: Dongjoo Lee, Jeongbin Park, Seungho Cook, Seongjin Yoo, Daeseung Lee, Hongyoon Choi. IAMSAM: Image-based analysis of molecular signatures using the Segment-anything model - Integrative analysis tool for tumor microenvironment [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 7431.